KFold Cross-Validation

[simon-ami]

What is the motivation for this task?

1. I would like to see support for k-fold cross-validation or even stratified k-fold cross validation
2. This would have multiple advantages, such as:
   • Less biased estimation of model skill
   • Utilization of limited data
   • Flexible trade-off between bias and variance

Describe the solution you'd like

(Stratified) K-fold cross validation being directly implemented in Anomalib dataset creation and training.

Additional context

I would also be thankful for an advice on a temporary workaround!

[samet-akcay]

thanks @SimonB97. We'll discuss this

[simon-ami]

Nice!
Could you maybe give a short explanation of how the splitting paramteres currently affect the dataset creation, or provide a link for further reading? In the meantime, i need some workaround to achieve cross validation. But since the splitting is configured in the config, i'm unable to wrap my head around how to do this..

Thanks!